Dynamic Label Injection for Imbalanced Industrial Defect Segmentation (ECCV 2024 Workshops)

This is the official implementation of the paper Dynamic Label Injection for Imbalanced Industrial Defect Segmentation.

Installation

Create your environment (Conda or Venv) and install the requirements with the following command:

pip install -r requirements.txt

Create MT dataset splits

python src/dsets/MT_gen_split.py --seed 42
                                 --dset_csv ./data/MT_dset/MT_dset.csv
                                 --output_dir ./data/MT_dset/splits

Launch training

python train.py --seed 42
                --config ./configs/MT/hyp.yaml 
                --model [timm-resnest50d,resnet18,mobileone_s1]
                --method [baseline,focal,balanced,wce,dli-cp,dli-p,dli-hh]
                --dli # use it when launching one of dli
                --poisson_prob [0,0.5,1.0] # 0.5 with dli-hh, 1.0 with dli-p, 0 otherwise
                --data_perc [0.1,0.25,0.5,0.75,1.0] # percentage of data in the training set
                --log_every 1 # saving weights after log_every epochs

Citation

Please cite with the following BibTeX:

@inproceedings{caruso2024dynamic,
  title={Dynamic Label Injection for Imbalanced Industrial Defect Segmentation},
  author={Caruso, Emanuele and Pelosin, Francesco and Simoni, Alessandro and Boschetti, Marco},
  booktitle={Proceedings of the European Conference on Computer Vision Workshops},
  year={2024}
}